Innovations in EDA Webcast: Measurement-based FET modeling using Artificial Neural Networks (ANNs)
1 小时 | 网上直播 -- 已存档的 | 地点和时间
Why this webcast is important:
Measurement-based transistor models offer an attractive alternative to conventional physically-based or empirical compact models of transistors. All devices that can be represented by the same equivalent circuit topology can be modeled by a single approach based on measuring and mathematically transforming the response data into the relevant nonlinear constitutive relations that define the elements in the equivalent circuit. In addition to maximum accuracy, this approach is quite general. It eliminates the need to develop device-specific model equations for each different device type (e.g. Si MOSFET and GaAs pHEMT), and reduces the need for multiple models that differ only in their closed-form empirical equations (e.g. Angelov, Advanced Curtice, TOM4 etc. for GaAs). The conventional approach could take a Ph.D. years to develop and implement a new model, and days for an expert to extract the parameters.
This talk introduces NeuroFET, a newly available complete measurement-based FET modeling flow, from automatically controlled data acquisition, ANN training, and non quasi-static FET model simulation for nonlinear circuit design. The flow, including automated measurement control and advanced neural network training and modeling capability, is built into Agilent ICCAP SW. The NeuroFET model is compiled into Agilent ADS. Innovative technical breakthroughs enable NeuroFET to optimally construct the complicated device charge functions and other state functions from knowledge only of the S-parameter measurements, resulting in excellent predictions at high frequencies for key effects such as AM-PM and ACPR distortion in power amplifiers. The inherent smoothness of ANNs provide much better dynamic range simulations compared to table-based approaches starting from the same data. Several additional advantages will be discussed. Results and validation on real devices will be presented.
Who should attend:
Device modeling engineers developing and/or extracting models for RF MESFET and HEMT devices and RF designers interested into predicting linear and distortion performance of circuits like amplifiers, mixers and switches by using a more accurate FET model than existing compact models.
|免费||At Your PC||Enroll to view the Feb 2, 2012 broadcast|